An Evaluation of Hubness Reduction Methods for Entity Alignment with Knowledge Graph Embeddings

Daniel Obraczka, Erhard Rahm

2021

Abstract

The heterogeneity of Knowledge Graphs is problematic for conventional data integration frameworks. A possible solution to this issue is using Knowledge Graph Embeddings (KGEs) to encode entities into a lower-dimensional embedding space. However, recent findings suggest that KGEs suffer from the so-called hubness phenomenon. A dataset that suffers from hubness has a few popular entities that are nearest neighbors of a highly disproportionate amount of other entities. Because the calculation of nearest neighbors is an integral part of entity alignment with KGEs, hubness reduces the accuracy of the matching result. We therefore investigate a variety of hubness reduction techniques and utilize approximate nearest neighbor (ANN) approaches to offset the increase in time complexity stemming from the hubness reduction. Our results suggest, that hubness reduction in combination with ANN techniques improves the quality of nearest neighbor results significantly compared to using no hubness reduction and exact nearest neighbor approaches. Furthermore, this advantage comes without losing the speed advantage of ANNs on large datasets.

Paper Citation

in Harvard Style

Obraczka D. and Rahm E. (2021). An Evaluation of Hubness Reduction Methods for Entity Alignment with Knowledge Graph Embeddings. In Proceedings of the 13th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2021) - Volume 2: KEOD; ISBN 978-989-758-533-3, SciTePress, pages 28-39. DOI: 10.5220/0010646400003064

in Bibtex Style

@conference{keod21,
author={Daniel Obraczka and Erhard Rahm},
title={An Evaluation of Hubness Reduction Methods for Entity Alignment with Knowledge Graph Embeddings},
booktitle={Proceedings of the 13th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2021) - Volume 2: KEOD},
year={2021},
pages={28-39},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010646400003064},
isbn={978-989-758-533-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 13th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2021) - Volume 2: KEOD
TI - An Evaluation of Hubness Reduction Methods for Entity Alignment with Knowledge Graph Embeddings
SN - 978-989-758-533-3
AU - Obraczka D.
AU - Rahm E.
PY - 2021
SP - 28
EP - 39
DO - 10.5220/0010646400003064
PB - SciTePress